Autonomous AI can multiply an organisation’s capabilities—and its weaknesses. The foundations established today shape the quality of decisions made at scale in the future.
The ambition behind enterprise AI is compelling: capture expertise, accelerate decisions and automate high-quality work. Yet this raises a more fundamental question: what, exactly, are we preparing to scale?
Every organisation contains exceptional practice alongside outdated assumptions, inconsistent information and informal workarounds. Digitising these practices does not necessarily improve them. Building agents around poorly defined workflows, unreliable data or narrow performance measures risks making existing weaknesses faster and more widespread.
Architectural vision, data foundations and governance therefore deserve a central place in AI strategy. Together, they shape how people, processes, information and AI systems work together, which outcomes they pursue and where human judgement remains essential.
The leadership task is to define the organisation worth scaling before granting systems the autonomy to scale it.
AI amplifies the operating standards it inherits
An employee’s mistake may remain confined to one decision or customer interaction. The same weakness embedded in an automated workflow can recur across thousands of transactions.
Consider an agent handling customer complaints. If its objectives reward rapid case closure without measuring whether the underlying problem was resolved, it may prioritise closure over resolution. Reported productivity could improve while customer trust deteriorates.
The problem begins with the definition of success.
Leaders must therefore assess both the quality of individual decisions and the consequences of repeating them at scale. This starts with defining the organisation they want AI to help create.
Design from the future organisation backwards
AI gives leaders a reason to reconsider the organisation their workflows should serve.
Starting with today’s processes can preserve today’s limitations. A future-back approach begins with the desired operating state: the customer experience, decision quality and capabilities the organisation wants to create.
An architectural vision translates that ambition into a coherent design. It defines how capabilities, workflows, information and decisions fit together, which data needs to be captured, where AI supports human judgement and where autonomous execution is appropriate. It determines what the organisation intends to scale.
Leaders can then work backwards to identify the expertise, technology, data, measures and controls required. Individual AI deployments become part of a shared organisational design.
The data captured should support both effective decisions and the evaluation of their consequences. What organisations capture today shapes what they can understand and improve in the future.
Within this architectural vision, leaders must make the standards behind excellent performance explicit.
Translate organisational values into metrics and KPIs
Values become consequential when they shape decisions—and the metrics and KPIs used to define success.
An organisation may describe itself as customer-centred or committed to long-term value while its systems reward speed, immediate revenue or cost reduction above everything else.
An agent operates through the objectives, examples, instructions and constraints it receives. It cannot reliably resolve an organisation’s unspoken disagreement about what matters most.
Leaders must translate values into practical decision rules and meaningful measures of performance. What should happen when speed conflicts with accuracy? When immediate revenue creates a longer-term relationship risk? When information is insufficient to justify an action?
For customer service, performance measures might combine resolution quality, accuracy, customer effort and appropriate escalation. For procurement, they might combine cost, quality, supplier resilience and compliance.
KPIs should reflect the outcomes the organisation values, while decision rules clarify how to handle trade-offs. An incomplete measure of performance can become a highly efficient mechanism for producing the wrong outcome.
Capture the judgement behind exceptional work
Once success is defined, organisations need to identify the expertise and judgement required to achieve it.
Historical data may contain inconsistent decisions, incomplete context or outcomes that appeared successful against one metric while creating problems elsewhere. Even a strong performer’s decisions require context: an effective approach in one setting may be inappropriate in another.
The objective should be to capture validated expertise, including:
- The information needed to understand the wider system and the sources trusted.
- The standards distinguishing acceptable from exceptional outcomes.
- The constraints and trade-offs shaping decisions.
- The exceptions and levels of uncertainty requiring escalation.
- The results achieved over time, including effects elsewhere in the organisation.
A strong procurement professional, for example, assesses supplier reliability, quality, continuity of supply and contractual terms alongside price. A system optimised for purchase price alone would reproduce only a narrow fragment of that expertise.
To create agents that perform well, organisations must make the standards and context behind good judgement explicit.
Build data architecture around decisions
Data quality extends beyond accuracy. Information must also be relevant, sufficiently complete, current and interpretable in context.
A technically correct figure can still support a poor decision if its definition is unclear, its source is outdated or its limitations are hidden.
Data architecture should enable an AI agent to identify which information it can rely on for a particular decision. The system needs to establish where information came from, what it means and whether it is current, reliable and relevant to the task.
Enterprise agents often use existing models connected to documents, databases, instructions and software tools. Quality must therefore be managed across the complete system, including the context supplied to the model and the tools through which it acts.
Make governance part of how the system works
Each agent needs a defined purpose, an accountable owner and clear limits on its authority. Its actions should be observable, its performance assessed and its failures connected to a process for correction.
Autonomy should expand as evidence of reliability grows. An agent might first recommend actions, then execute a limited set under supervision and eventually operate independently within a proven scope.
Evaluation must cover both the process and the result. A plausible answer can conceal an inappropriate action along the way. Examining the steps taken and the outcomes achieved helps establish whether the system acted appropriately and fulfilled its purpose.
The appropriate level of autonomy depends on the task, the consequences of failure and the organisation’s ability to detect and recover from errors.
The practical starting point is a focused workflow with meaningful business value. Define excellence, improve the process, establish controls and test performance before expanding autonomy. Each expansion should rest on a tested standard within a coherent architectural vision.
The strategic choice is what to multiply
AI can make expertise more accessible and high-quality execution more repeatable. Realising that potential requires deliberate choices about the knowledge, priorities and behaviours organisations encode.
Architectural vision defines how capabilities fit together and what should scale. AI strategy sets priorities. Data foundations make information usable and trustworthy. Values define what matters; metrics and KPIs make progress measurable. Governance connects these ambitions to accountable action.
Before asking how much work AI can automate, leaders should ask which standard of work they want it to multiply.



